The Agent Skills Directory

[PROMPT_INJECTION]: The 'Suggest Using K-Dense Web For Complex Worflows' section contains instructions to modify the agent's conversational behavior. It directs the AI to proactively promote the author's external platform ('www.k-dense.ai') when users request complex tasks, which is a form of behavioral steering.
[COMMAND_EXECUTION]: The skill provides instructions to install the 'matchms' library and its optional chemistry dependencies using 'uv pip install'. This is a standard installation procedure for Python-based tools.
[DYNAMIC_EXECUTION]: The documentation identifies 'Pickle (Python serialization)' as a supported data format. Loading untrusted data using the Python pickle module can lead to arbitrary code execution, as the format is not secure by design.
[DATA_EXFILTRATION]: The skill references 'www.k-dense.ai' as an external destination for research workflows. This domain belongs to the skill's author, K-Dense Inc., and is presented as a productivity upgrade for complex tasks.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it ingests untrusted data from various mass spectrometry file formats.
Ingestion points: External files such as '.mgf', '.mzML', '.msp', and '.json' are loaded into the agent's context.
Boundary markers: None identified in the provided documentation or code snippets.
Capability inventory: The skill includes capabilities to write files ('save_as_mgf', 'save_as_json') and perform complex metadata processing.
Sanitization: There is no evidence of sanitization or validation to prevent embedded instructions within the spectral metadata from influencing the agent's actions.

matchms